Professional Documents
Culture Documents
1. Introduction
In recent years a number of studies [2-4] have been carried out to perform the sensitivity
analysis of fuel cell parameters. Among these studies, the one done by Grujicic et al.
(2003) [2] is very comprehensive, which focused on optimization of operating parameters
1
Corresponding Authors: wang@oakland.edu; 248-370-2224
25
ECS Transactions, 33 (21) 25-37 (2011)
and dimensions of fuel cells. The focus of this paper is to investigate how the fuel cell
performance is sensitive to some gas diffusion layer parameters.
The gas diffusion layer (GDL), typically made of carbon paper or carbon cloth, is an
important component of proton exchange membrane(PEM) fuel cells. The GDL provides
mechanical support to catalyst layers and membrane; it conducts heat and electricity, and
carries reactants to the catalyst layer; more importantly, it delivered reactant gas to the
reaction sites, and take the product water away from the reaction sites. Therefore, the
GDL plays an important role in maintaining the effective water management in PEM fuel
cells. The GDL properties such as its porosity, permeability, thickness and thermal
conductivity significantly affect the water management in the fuel cell, whereas electrical
conductivity influences transmission of electrons to and from bipolar plates.
From last few years significant amount of research has been done for developing models
for PEMFC. These modeling works have been reviewed by several researchers [5-10].
For the simplification purpose, most of the earlier fuel cell models were assuming the
isotropic GDL structure. However, several experimental investigations (11-14) show that
the gas diffusion layer is highly anisotropic. For example, the in-plane permeability of
GDL is approximately twice as high as the through-plane permeability (11-12). Nitta et
al. (14) measured both the in-plane and through-plane electrical conductivity, and found
that the electric conductivity in the in-plane direction of the measured GDL is three to
four times higher than that in the through plane direction. The GDL thermal conductivity
is expected to exhibit the same degree of anisotropy (5). Most recently, a few research
groups have accounted for the anisotropic nature of GDL structure in the modeling (15-
16). The degree of the anisotropic property of GDL on the performance of fuel cells, is
however, not quantified.
Therefore, the objective of this paper is to study the effect of the key GDL parameters
like porosity, permeability, thermal conductivity and electrical conductivity on the fuel
cell performance by applying the sensitivity analysis to a non-isothermal fuel cell model.
The analysis of variance, commonly known as ANOVA, helps in determining the
contribution of each factor to the total variation from the overall mean value [17].
Genichi Taguchi has pioneered this method in engineering application [18]. The ANOVA
method will be implemented in this study to do the sensitivity analysis of the key GDL
parameters on fuel cell performance.
2. Mathematical Model
26
ECS Transactions, 33 (21) 25-37 (2011)
Figure 1 shows the schematic sketch of modeling geometry consisting of a cathode and
anode electrode with a proton exchange membrane (PEM) as the electrolyte sandwiched
in between. Each electrode consists of a gas diffusion layer (GDL) and a catalyst layer
which exists at the interface of gas diffusion layer and membrane. It is assumed that all
the parameters change in X and Y directions only.
A
Catalyst
e-
GDL GDL
Anode Cathode
H+
PEM
Y
The steady state, single phase and non-isothermal model include the following governing
equations:
In a PEM fuel cell, the current can be split into two parts: ionic current and electronic
current. Protons travel through the membrane to form an ionic current denoted by ie,
while electrons transfer through only the solid matrix of electrodes to form an electronic
current denoted by is. Using Ohm’s law, the current can be presented in terms of
potential given as
27
ECS Transactions, 33 (21) 25-37 (2011)
current source term, A u m -3 ; the subscript s denotes the property of solid phase and e
denotes that of electrolyte phase. The source terms are given in equations (3, 4), which
are the results of electrochemical reaction occurring at anode and cathode catalyst layers.
The agglomerate model describes the current density in an active layer consist of
agglomerate of ionic conductor material and electrically conducting particle covered
partially with catalyst. The agglomerate model describes the transfer current density as
following [19-20].
Anode:
6(1 H )FDHagg § agg ref § 2F · · § jo,a s jo,a s ·
ja ¨ cH cH exp ¨ K ¸ ¸ ¨1 Ragg coth Ragg ¸ (5)
© RT ¹ ¹ ¨© ¸
agg 2 ref agg ref agg
(R ) © aFcH DH aFcH DH ¹
Cathode:
12(1 H ) FDOagg agg § jo,c s( R agg )2 § 0.5F · ·¸ j s( R agg )2 § 0.5F · (6)
jc R cO ¨1 exp ¨ K ¸ coth o,c ref agg exp ¨ K¸
( R agg )2 ¨ 4FcOref DOagg © RT ¹ ¸¹ 4 FcO DO © RT ¹
©
where c agg is the gas concentration at the surface of the agglomerates, mol m 3 ; c ref is the
dissolved gas concentration at the reference state, mol m 3 ; F is the Faraday constant,
A s m 1 ; D agg is the diffusion coefficient of the dissolved gas inside the
agglomerate, m 2 s 1 ; R agg is the agglomerate radius, m ; j0 is the exchange current
density, A m 2 ; s is the specific surface area of the catalyst layer, m 2 m 3 .
Henry’s law describes the dissolved gas concentration at the surface of the agglomerates
which is related to the molar fraction of the respective species.
28
ECS Transactions, 33 (21) 25-37 (2011)
P
c agg cgas (10)
H
where H is the Henry’s constant Pa.m3/mol.
The potential at the anode is arbitrarily chosen to be zero, and the total cell voltage is
used as boundary condition at the cathode. All of the remaining boundaries are treated as
insulation.
Gas diffusion layers and catalyst layers are porous media. Darcy’s law for flow through
porous media and mass conservation equation are used to describe the velocity
distribution.
Continuity equation:
Uu S (11)
where u is the gas velocity (m/s), is the mixture density of the gas phase (kg/m3),
which is calculated as
§ ·
U
P ¨
RT ¨¨ ¦xi M i ¸
¸¸ (12)
© i ¹
where R is the universal gas constant 8.314(J/mol.K), T is the temperature (K), Mi is the
molar mass (kg/mol), and xi is the mole fraction. The subscript i (or j) represents each
species of hydrogen and water on the anode side, and oxygen, water, nitrogen on the
cathode side. Inlet pressure Pa, in and Pc, in are laid as a boundary condition on anode and
cathode side respectively.
The source term S in equation (11) accounts for the total consumption and production
during the electrochemical reaction. In the catalyst layer the reaction rate Ri
corresponding to each species is given by,
j
RH 2 a M H 2 (13)
2F
j
RO 2 c M O 2 (14)
4F
jc
RH 2O M H 2O (15)
2F
The feed-gas mass fractions are specified at the inlet of anode and cathode.
Darcy’s Law:
G N
u p (16)
P
29
ECS Transactions, 33 (21) 25-37 (2011)
where N denotes the effective permeability tensor (m2) for electrode, represents the
dynamics gas viscosity (Pa.s), and p is pressure (Pa).
¦ D { MM M P G
[ UZi ij ( Z j Z j ) (x j Z j ) } Zi U u ] Ri (17)
j M P
j 1
where Dij is the diffusion coefficient (m2/s); Ri the reaction rate (kg/ (m3.s); x, the mole
fraction; , the mass fraction; and M is the molecular mass (kg/mol). The density is the
mixture density as given by the equation (12). On the cathode side transportation
equation is solved for oxygen and water and the third species, nitrogen is obtained by
mass balance.
wN 2 1 wO 2 wH 2O (18)
Whereas on the anode side, the transportation equation is solved for hydrogen and water
is obtained from the mass balance.
wH 2O 1 wH 2 (19)
All the governing equations were solved using COMSOL Multiphysics commercial
software. This software is based on the finite element technique. The COMSOL chemical
engineering module is referred for setting up the problem in COMSOL. The model setup
is similar to the one used by Shi & Wang (2007) [20], except that a non-isothermal model
is formulated in this study as against an isothermal model. The stationary non-linear
30
ECS Transactions, 33 (21) 25-37 (2011)
solver is used till the model is converged; afterward the parametric non-linear solver is
used to compute the results at different values of cell voltage.
Taguchi method will be used to perform sensitivity analysis, and this methodology relies
more on engineering judgment than absolute statistical values [17]. The fuel cell
performance is usually characterized by current density at a given voltage or power
density. Different parameters like geometry, material properties, operating parameters are
used as inputs for the model. During the course of sensitivity analysis, these parameters
are called as factors per Taguchi terminology. These factors are assigned discrete values
called as levels. These levels are distributed equally across the mean value of each factor.
Gas diffusion layer (GDL) properties are chosen as factors in the following sensitivity
analysis. Table 1 shows seven different parameters (GDL properties) used for analysis.
Three different levels are considered for these parameters, with level 2 as base level or
mean. Level 1 and level 3 are assumed at 10% on the either side of mean.
As the number of factors considered for sensitivity analysis increases, the experimental
work becomes unexceptionally high. To reduce the number of analyses, the orthogonal
matrix method is used. This method contains columns for each factor and rows for
analyses; each analysis represents a particular combination of the levels of each factor.
Some standard orthogonal matrices are available to accommodate specific number of
factors. In this study a standard L18 matrix (22-23) is used. Eighteen different analyses
were carried out and the current density at the voltage of 0.6 V for each analysis is
calculated. Table 2 represents the L18 matrix used in this study. As shown in Table 2, a
unique combination of different levels of factor is used in each analysis. Each level
repeats equal number of times (six times) for each factor in the matrix. The values of the
objective function (current density) are in the last column for each analysis, and the
overall mean is at the bottom of last column.
31
ECS Transactions, 33 (21) 25-37 (2011)
The deviation caused by the level of each factor is calculated by subtracting the overall
mean value from this mean of each level of each factor. Once this deviation for each level
of each factor is determined, then the deviation value is squared and the sum of all such
squares is calculated. The percentage of this sum of a given factor to the cumulative sum
of all the factors gives the relative importance of that factor. This value, however, is
without accounting for the error associated due to the linear superimposition assumption
(17, 23).
Once the sum of square due to the error and the corresponding number DOF for the error
is calculated, the error variance and variance ratio can be calculated [17] as
32
ECS Transactions, 33 (21) 25-37 (2011)
Table 3 presented the intermediate and final results of sensitivity analysis. The variance
ratio results are displayed in the last column. Porosity (factor A), followed by through
plane permeability (factor C), shows the greatest impact on the current density for a fixed
cell voltage of 0.6 V. All other GDL properties don’t present significant effect on the
performance. The variance ratios also show the effect of in-plane and through-plane
thermal conductivity (factors D and E) and electrical conductivity (factors F and G), in
both cases through-plane (E and G) has got relatively more impact on the performance
than the in-plane (D and F).
From the result of eighteen analyses in Table 2, two extreme cases with the highest
(analysis #13) and lowest values (analysis #1) of current density were chosen for
comparison. Figure 2 shows the plot for temperature distribution. Figure 3 and Figure 4
show the oxygen concentration and water concentration comparison, respectively. Figure
2 that higher current density results in higher temperature distribution. Meanwhile the
oxygen concentration from inlet to outlet of the flow channel is depleted faster for higher
current density operation as shown in Figure 3. More water is therefore produced at
higher current density as shown in Figure 4.
33
Figure 2: Temperature distribution comparison within a fuel cell.
353.2 353.2
353.6 353.6
353.6
354 354.4
354.4
354.4
356.4 356
355.6
355.6 355.6 355.2 355.2
355.6 355.2
355.2
354
354 354
353.2
34
353.2
354.4 354
354.8 354.8 354.8
355.6 355.2
(a) analysis #1
.6
355
355.6
355.6
355.2
354.8
353.6 353.6 353.6
3 54 353.2 353.2
353.2
ECS Transactions, 33 (21) 25-37 (2011)
0.17
0.15
0.19
0.195
0.2
0.19
0.1
0.13
7
0.15
0.11
0.165
0.16
0.1
45
0.03
35
ECS Transactions, 33 (21) 25-37 (2011)
0.22
0.205
0.21
0.26
0.17
0.24
0.21 0.22
0.26 0 27
0.
22
5
0.2
3
0.2
4
0.24
0.34
0.29
0.27
0.2
55
0.265
0.
0.33
27
4. Conclusions
36
ECS Transactions, 33 (21) 25-37 (2011)
4 It is observed from the variance ratio of both thermal conductivity and electrical
conductivity, that through-plane conductivities are more significant than in-plane
conductivities.
6. References
37